Skipping over empty columns in sparse matrix

Hello, iglers,

The code for the mosek quadprog wrapper crashes if there are empty columns at the end of the constraint matrix.

https://github.com/libigl/libigl/blob/e2aa034ab822d492ba74d736a8d823a317e8520e/include/igl/mosek/mosek_quadprog.cpp#L175-L180

For example, let's say that the constraint matrix is

A = [1 0]

Then,

Av = [1]
Ari = [0]
Acp = [0 1 1]

When j=1, the code tries to dereference Av[1] and Ari[1], which causes a crash on our platform (Windows 10 & Visual Studio 2017). A simple fix is to check if the nnz in the jth column is 0.

According to the Mosek API page, MSK_putacol() resets all the elements in column j to zero and then..." So, it might still be a good idea to call this function with a zero column to make sure that the column is actually zeroed out.

    int nnz = Acp[j+1] - Acp[j]; // number of nonzeros in jth col
    if(nnz > 0)
    {
      mosek_guarded(
        MSK_putacol(
          task,
          j,
          nnz,
          &Ari[Acp[j]],
          &Av[Acp[j]])
        );
    }else
    {
      mosek_guarded(
        MSK_putacol(
          task,
          j,
          0,
          0,
          0)
        );
    }

Start in include/igl/mosek/mosek_quadprog.cpp around lines 175-180 and inspect how sparse column ranges are passed to MSK_putacol(). Reproduce the failure with a constraint matrix containing an empty trailing column, then verify that the wrapper handles the zero-length column without dereferencing invalid entries and preserves the intended zero-column state.